References

All source documents, links, and reading material referenced across the engagement.

Claude Code & Claude Agent SDK

• Claude Code — Anthropic’s official CLI for Claude
• Claude Agent SDK overview
• Building effective agents — Anthropic research
• Tool use
• Prompt caching
• Extended thinking

Model Context Protocol (MCP)

• MCP — Introduction
• MCP — Specification
• MCP — Server quickstart

LIFEOSAI

• LIFEOSAI — multi-agent orchestration substrate. Coordinates agent wakeups, task handoffs, skills, memory, governance, and budgets. Runs 40+ agents across multiple companies in production.
• Architecture: Postgres-backed orchestration runtime with heartbeat-based wakeups, adapter engine (Claude, OSS-80B, OpenAI), and an office library for institutional memory.
• The workshop uses LIFEOSAI as the substrate for the 9-agent AI Guardrail Lab.

Paperclip (OSS upstream)

• Paperclip on GitHub — open-source multi-agent orchestration framework (MIT licence)
• Same pattern as the LIFEOSAI orchestration engine, available for anyone to use and extend
• Features: org charts, budgets, governance, bring-your-own-agent (BYOA)

AI Incident & Guardrail References

• OECD AI Incidents Monitor — global AI incident tracking
• AIAAIC Repository — AI and algorithmic incident cases
• Stanford AI Index — annual AI measurement and tracking

Agent AI safety & guardrails

• Anthropic — Agent safety best practices
• Anthropic — Responsible Scaling Policy
• NIST — AI Risk Management Framework
• OWASP — Top 10 for LLM Applications

MetaWeave

• MetaWeave HQ — GitHub
• Agent Companies framework — agentcompanies/v1 schema for defining multi-agent companies

Engagement artefacts

• Proposal summary — agreed scope for this engagement
• Workshop site: nyk-workshop.metaweave.in
• Slide decks, lab repos, and post-engagement deliverables: links added during the workshop

Glossary

Term	Definition
Agent	An LLM-driven system that observes, reasons, and acts via tools in a loop until a goal is met or a stop condition fires.
Multi-agent	Multiple agents coordinated by an orchestration layer — each with its own role, skills, and instructions.
Substrate	The orchestration runtime that makes agents coordinate — wakeups, handoffs, memory, governance. LIFEOSAI is the substrate.
Office	A company-level workspace in LIFEOSAI containing a team of agents, their tasks, skills, and shared memory. The AI Guardrail Lab is one office.
Guardrail	Technical, process, and organisational controls that bound the behaviour of an agent system in production.
Fit-Gap	Structured comparison of an existing system against a reference target (here: modern application architecture + agent-native patterns).
Claude Code	Anthropic’s CLI for Claude — an agent itself that reads, writes, and runs code.
Claude Agent SDK	Anthropic’s SDK for building custom agents with Claude — handles the agent loop, tools, and orchestration.
MCP	Model Context Protocol — open standard for connecting agents to external tools, data, and systems.
Paperclip	Open-source upstream of the LIFEOSAI orchestration pattern (MIT licence).
DX Promotion Group	NBS internal group responsible for digital transformation budget and approvals.